Mathematicians at RUDN University have conducted a comparative study on forecasting models for 5G and 6G networks, revealing their strengths and weaknesses. These networks will need to track indicators and predict them accurately to make decisions about network division and load balancing. To achieve this, machine learning models are commonly used for prediction.
In the research published in the journal Future Internet, the mathematicians compared two forecasting models: the seasonal integrated autoregressive moving average (SARIMA) model and the Holt-Winter model. They utilized data from a Portuguese mobile operator, specifically on traffic volumes for downloading and uploading over fixed periods.
Both models prove to be suitable for predicting traffic for the next hour. However, SARIMA performs better in predicting traffic from the user to the base station, with an average error of 11.2%, which is 4% lower than the second model. On the other hand, the Holt-Winter model is more effective in predicting traffic from the base station to the user, with an error of 4.17% instead of 9.9%.
According to Irina Kochetkova, Ph.D., Associate Professor at the RUDN Institute of Computer Science and Telecommunications, 5G and 6G networks will support advanced technologies such as drones, virtual and augmented reality. The increased number of connected devices leads to a surge in traffic, resulting in network congestion, reduced service quality, network delays, and data loss. Therefore, the network architecture must adapt to varying traffic volumes and support different types of traffic with distinct requirements.
While both SARIMA and Holt-Winter models are effective in predicting traffic averages, their suitability depends on the specific dataset. Associate Professor Kochetkova believes that future research should focus on combining statistical models with machine learning methods to enhance accuracy and detect anomalies.
This study provides valuable insights into forecasting models for 5G and 6G networks, highlighting the strengths and weaknesses of two commonly used models. By understanding their limitations and potential, researchers and industry professionals can work towards developing more robust and accurate prediction models that will enable efficient network management in the future.
